J0nasW/science-datalake

Science Data Lake (release 2026.10)

A DOI-linked, source-preserving integration of open scholarly metadata. Six open sources are linked through normalised DOIs while every source keeps its own schema and values, so that measurements from different sources (three citation counts, two retraction signals) can be compared directly.

• Release: 2026.10 (tag v2026.10), DOI 10.57967/hf/XXXX
• Licence: CC BY-NC-SA 4.0 for the whole dataset (see LICENSE; component sources and their licences in ATTRIBUTION.md)
• Size: 572 Parquet files, 619 GB; MANIFEST.sha256 lists the SHA-256 of every data file
• Views database: datalake.duckdb (142 views in 17 schemas, no data)
• Paper: Wilinski, J. Science Data Lake: a DOI-linked, source-preserving integration of open scholarly metadata (Scientific Data, under review)

Contents

Schema	Source (version)	Principal table rows	Upstream licence
openalex	OpenAlex (snapshot 2026-02-03)	510,372,821 works	CC0 1.0
sciscinet	SciSciNet v2	249,803,279 papers	BSD-3-Clause-Clear
crossref	Crossref public data file 2026-03 (no abstracts)	179,536,204 works	CC0
retwatch	Retraction Watch	71,106 notices	CC0
ros	Reliance on Science v64	47,844,586 patent-to-paper citations	CC BY-NC 4.0
p2p	PreprintToPaper 2.0.0 (no abstracts)	145,517 preprints	CC BY 4.0
10 ontology schemas	AGROVOC, ChEBI, CSO, DOID, GO, MeSH, MSC2020, NCIT, PhySH, STW	1,266,884 terms	see ATTRIBUTION.md
xref	derived in this work	315,754,730 rows in unified_papers	CC BY-NC-SA 4.0

MeSH data are provided by the U.S. National Library of Medicine and may not reflect the most current MeSH release.

xref.unified_papers has one row per normalised DOI found in OpenAlex, SciSciNet or Retraction Watch (32 columns), with the citation counts of OpenAlex, SciSciNet and Crossref, both retraction signals, SciSciNet indicators, patent citations and preprint links side by side, plus coverage flags. xref.topic_ontology_map holds 15,336 candidate links between OpenAlex topics and ontology concepts ranked by label similarity (auxiliary; not curated equivalence mappings).

Query without downloading

-- authenticate (token from `hf auth login`); anonymous reads are rate-limited
CREATE SECRET hf (TYPE huggingface, PROVIDER credential_chain);
ATTACH 'hf://datasets/J0nasW/science-datalake@v2026.10/datalake.duckdb' AS sdl;
USE sdl;
SELECT count(*) FROM xref.unified_papers WHERE has_crossref AND has_sciscinet;

Selective queries work well remotely; for full-table analyses download the release (huggingface-cli download J0nasW/science-datalake --repo-type dataset --revision v2026.10) and verify it with sha256sum -c MANIFEST.sha256.

Citation

@misc{wilinski2026sdl,
  author = {Wilinski, Jonas},
  title  = {Science Data Lake (release 2026.10)},
  year   = {2026},
  publisher = {Hugging Face},
  doi    = {10.57967/hf/XXXX}
}

Please also cite the component sources listed in ATTRIBUTION.md.